The surface energy budget is an accounting problem
Bowen-ratio energy balance is a way to understand how available energy near Earth’s surface is divided between warming the air and evaporating or transpiring water. It is useful in studies of wetlands, crops, bare ground and open water, but a reported evapotranspiration number is not a direct reading from a bucket of vanished water. The result is inferred from measured radiation, ground heat, temperature-humidity gradients and assumptions about turbulent transfer. Translating the method well means making that energy accounting explicit and checking which terms and signs the researcher has used.
In a simplified near-surface balance, net radiation Rn is apportioned to sensible heat flux H, latent heat flux LE and ground or storage heat flux G, with sign conventions stated. A common daytime teaching form is Rn minus G equals H plus LE. All fluxes are rates of energy per horizontal area, typically watts per square metre. The equation is a conservation statement under its modelling assumptions. It does not guarantee that every field sensor provides a perfect measurement of each term or that neglected heat storage and advection are genuinely zero.
Evapotranspiration combines evaporation from soil, water or wet surfaces with water movement through vegetation followed by transpiration. Latent heat is the energy associated with a phase change of water, while sensible heat changes temperature without that phase change. If incoming energy largely evaporates water, LE can dominate; if a dry surface cannot supply much water, H may dominate instead. This intuitive picture is only a starting hypothesis until quantified fluxes and site conditions support it.
The Bowen ratio and what its denominator means
The Bowen ratio β is defined as sensible heat flux divided by latent heat flux, H divided by LE, under a consistent sign convention. A β of 0.5 means the sensible heat flux is half the latent heat flux in that interval. A β of 2 means sensible heat is twice the latent component. It does not mean two degrees, two millimetres, a 200-percent relative humidity or that exactly two water molecules are evaporated for each unit of atmospheric warming. Because it is a ratio of two quantities with matching energy-flux units, β is dimensionless.
Suppose available energy Rn minus G is 300 W m−2 and β = 0.5. Since H = 0.5 LE, the balance becomes 300 = 1.5 LE, giving LE = 200 W m−2. Then H = 100 W m−2. The energy check is 200 plus 100 equals 300. This worked example illustrates why the ratio alone is insufficient to estimate evaporation: β determines the partition, while the available energy sets the overall magnitude. An identical β under 100 W m−2 available energy gives a much smaller latent flux.
A reader should ask whether the reported β is calculated from measured gradients, from flux measurements, or from a model. Also ask whether the underlying interval has sufficient energy and vapour gradients to produce a stable ratio. Dividing by an LE near zero can create a huge positive or negative β that tells more about numerical sensitivity than surface behaviour. A valid-looking ratio needs its numerator, denominator and measurement conditions, not merely an isolated decimal.
How heat and vapour gradients can estimate the ratio
The Bowen-ratio method often uses near-surface gradients in air temperature and vapour pressure measured at two heights. Under an assumption of comparable turbulent transfer behaviour for heat and water vapour, β can be related to the psychrometric constant γ multiplied by the temperature difference divided by the vapour-pressure difference. The exact sign depends on which height is subtracted from which, so both differences must use the same reference order. This formula connects observations from thermometers and humidity measurements to a ratio of turbulent fluxes.
The psychrometric constant is commonly expressed in kilopascals per degree Celsius. In a simplified illustrative computation, take γ = 0.066 kPa per degree Celsius, a temperature difference of 1.5 degrees Celsius and a vapour-pressure difference of 0.20 kPa in the consistent gradient direction. Then β = 0.066 times 1.5 divided by 0.20, which is 0.495. The result is dimensionless because multiplying γ by the temperature difference gives kilopascals, cancelling the kPa in the denominator.
Using the wrong vapour variable can break this calculation. Relative humidity is a percentage defined in relation to temperature-dependent saturation, not the vapour pressure in kilopascals used in the simple gradient equation. Before using relative humidity data, convert appropriately to actual vapour pressure at each measured air temperature. If a researcher simply divides a one-degree temperature difference by a ten-percentage-point relative-humidity difference, the resulting number is not the same Bowen ratio formula and cannot be interpreted as such.
From latent energy to water depth
Latent heat flux is an energy rate, not directly a rainfall or evaporation depth. To convert it to evaporated water mass, divide the integrated energy by the latent heat of vaporisation, which varies with temperature. A useful illustrative value near ordinary conditions is 2.45 million joules per kilogram, but applications should choose a value appropriate for the method and temperature. One kilogram of water per square metre corresponds approximately to one millimetre of water depth.
For example, let LE equal 200 W m−2 over one complete hour. A watt is one joule per second, so the hourly energy per square metre is 200 times 3,600, or 720,000 joules per square metre. Dividing by 2,450,000 joules per kilogram yields about 0.294 kilograms per square metre, approximately 0.294 mm of evapotranspiration. The one-hour interval and latent-heat value must be reported. Multiplying the 200 by twenty-four directly would have incorrect units for daily depth.
Now suppose LE is a true 24-hour mean of 70 W m−2. The integrated daily energy is 70 times 86,400 = 6,048,000 J m−2. At the same illustrative latent heat, this corresponds to about 2.47 mm of water per day. This calculation would be inappropriate if the 70 W m−2 value represented only midday conditions or a short sunny period. Evapotranspiration is calculated from time-integrated flux; a single midday snapshot is not a daily average.
Net radiation is an input, not synonymous with sunlight
Net radiation includes absorbed shortwave radiation and net longwave exchange. It can be smaller than incoming sunlight because some sunlight is reflected and thermal radiation leaves the surface. At night net radiation often becomes negative, while available energy may reflect heat released from storage or supplied by the air. Calling Rn ‘sunlight’ hides the distinction between gross incoming solar energy and the signed net energy exchange of the surface.
Ground heat flux G represents heat flowing into or out of soil or other storage. Its sign convention must be read from the source documentation. Many daytime formulations treat positive G as energy travelling down into the ground, reducing energy available to turbulent sensible and latent fluxes. In a wetland or water body, storage within the water column and other terms may be material. Ignoring them without justification can make apparently precise LE estimates systematically wrong.
A short exercise uses Rn = 450 W m−2, downward G = 50 W m−2 and β = 0.6, all following that convention. Available energy is 400 W m−2. LE is 400 divided by 1.6 = 250 W m−2, while H is 150 W m−2. The check is 50 + 150 + 250 = 450. If another paper defines upward G positive, a literal copy of this subtraction may be wrong. Reading the sign convention is as important as remembering the formula.
What a wetter surface can do to the flux partition
Water supply influences the available pathways for surface energy. With sufficient water and favourable atmospheric demand, evaporation and transpiration can consume a larger fraction of available energy. A relatively dry surface may put more energy into sensible heat. A lower Bowen ratio therefore can be consistent with greater relative evaporation, but the ratio is not an automatic soil-moisture sensor. Wind, radiation, plant conductance, air dryness and heat storage also influence the exchange.
Imagine two observed daytime intervals with the same available energy of 400 W m−2. For β = 0.25, LE = 320 and H = 80 W m−2. For β = 1.5, LE = 160 and H = 240 W m−2. These are mathematically valid partitions under the simple closure assumption. The larger latent flux in the first case reflects a greater energy fraction allocated to evaporation and transpiration; it does not prove that the water supply caused the difference without supporting environmental observations.
An article that reports a rising Bowen ratio through a drying season should examine whether the incoming energy changed, whether plants altered stomatal conductance, whether temperature-humidity gradients were measured at consistent heights and whether the instruments maintained calibration. Seasonal changes may be real, but the ratio can respond to several coupled mechanisms. Explaining those mechanisms is more useful than turning the ratio into a single causal label.
The problem of a Bowen ratio near minus one
Algebra gives LE = (Rn minus G) divided by (1 plus β). If β approaches minus one, the denominator approaches zero, so very small measurement errors can yield enormous positive or negative LE estimates. The warning is mathematical and physical: the gradient approach becomes numerically unstable under some conditions, especially when gradients are small, reversed or dominated by measurement noise. It is not sensible to publish an enormous evaporation value merely because a spreadsheet formula returns it.
Take an illustrative available-energy value of 100 W m−2 and β = −0.99. Blind division gives 100 divided by 0.01, or 10,000 W m−2, an implausibly large estimated latent flux for the simplified ordinary surface example. A tiny change in β to −1.01 reverses the sign of the computed flux. This sensitivity demands quality screening and physical investigation rather than accepting the algebraic result. The problem reflects an ill-conditioned inference, not a sudden release of physically unlimited energy.
Negative Bowen ratios can occur in real surface-exchange situations under particular flux directions, including some night-time and advective settings; a negative value is not automatically an error. The critical questions are which terms are positive under the adopted sign convention, whether the energy balance is complete, and whether the gradients are sufficiently resolved. A robust method describes when data should be excluded or interpreted separately, avoiding a simplistic rule that all negatives are invalid.
Sensor heights, fetch and a coherent air column
The Bowen-ratio gradient method relies on temperature and vapour-pressure measurements at two heights that sample comparable turbulent exchange over a meaningful underlying surface. If one sensor is shaded by foliage while another is in direct radiation, apparent temperature gradients may partly reflect instrument exposure rather than atmospheric heat transport. If a humidity sensor sits next to a wet structure, its vapour reading may be unrepresentative. Precision in a sensor readout does not repair a poor installation.
Fetch describes the upwind surface influencing the air sampled by a sensor. A station near the boundary of an irrigated field and dry soil may respond to a mixture of surfaces that varies as the wind direction changes. A simple two-height gradient may therefore not describe a single homogeneous surface. The investigator needs information about wind, sensor heights, surrounding land cover and whether the sampled footprint fits the question. Without it, a precise evaporation figure can be spatially misleading.
Height consistency is equally important during a long experiment. Vegetation grows, leaves emerge and canopies change aerodynamic properties; the distance between a sensor and the effective exchange surface may shift even if the instrument pole remains fixed. A seasonal comparison should document such changes and check that the measured gradients remain meaningful. Translating an instrument reading into a water-use claim without this context can hide the primary source of uncertainty.
Vapour-pressure differences near zero amplify noise
In β = γ ΔT divided by Δe, a small vapour-pressure difference Δe in the denominator makes β sensitive to small measurement errors. If the two humidity measurements differ by only 0.01 kPa, a sensor offset of several thousandths of a kilopascal can materially change the calculated ratio. Temperature sensors have analogous challenges when ΔT is small. Differential calibration often matters more than absolute accuracy because the method is built on differences between two nearby sampling levels.
Suppose γ = 0.066 kPa per degree Celsius and ΔT = 0.2 degrees. If Δe is 0.02 kPa, β = 0.66. If an unrecognised offset changes the apparent Δe to 0.01 kPa, β doubles to 1.32 without any change in the measured temperature difference. The example shows why frequent sensor comparisons and correction of differential biases matter. It does not claim that all small gradients are unusable; their uncertainty must be assessed relative to the actual signal.
A responsible workflow examines calibration swaps, ventilation, radiation shields, humidity equilibration time and consistency of sensor sampling. When uncertainty overwhelms the gradient, report the observation as unreliable rather than manufacturing a refined β through excessive decimal places. Missing or screened periods should remain identifiable in the final daily energy accounting. This principle transfers to any measurement based on the difference between two large, nearly equal quantities.
Daily and seasonal evaporation are not copies of a midday formula
Sunlight, ground heat, air temperature, vapour pressure and atmospheric turbulence change throughout the day, so Bowen-ratio values and latent flux can vary or change sign. A valid daily water depth requires integration over appropriate valid intervals, with clear rules for missing values and quality screening. Averaging β first and then applying it to averaged energy is not always equivalent to calculating interval fluxes and integrating, because the formula is nonlinear.
Imagine one hour has available energy of 400 W m−2 with β = 0.5, and a second hour has available energy of 100 W m−2 with β = 2. The respective latent fluxes are about 266.7 and 33.3 W m−2. Their two-hour energy-integrated latent contribution differs from any naive estimate obtained by plugging the simple average β of 1.25 into the simple average available energy of 250 W m−2. Correct computation respects each interval before summing energy.
Daily and seasonal estimates also require clear treatment of rain, dew, nighttime condensation and data gaps. A positive net evapotranspiration total may conceal intervals of negative flux under a signed convention. A reported annual total might represent only a growing season with gap filling or model assumptions. The reader should inspect observation coverage before comparing totals across landscapes or years.
Bowen ratio is not the same as an eddy-covariance measurement
Eddy covariance estimates turbulent fluxes from high-frequency covariation of vertical wind with temperature and water-vapour fluctuations. Bowen-ratio energy balance instead uses energy-budget measurements and an inferred sensible-to-latent partition based on gradients. Both methods can estimate surface exchange but rely on different instruments and assumptions. A report should not imply that a two-height temperature-humidity system directly measured eddy covariance simply because it produced a latent heat flux number.
The surface-energy balance itself is often imperfectly closed in field studies. Turbulent flux measurements may not sum exactly to measured available energy because of storage, advection, instrumental limitations, footprint differences and methodological uncertainty. Bowen-ratio methods may effectively impose an energy partition under the chosen balance, but that does not independently demonstrate every term is measured without error. A comparison with eddy-covariance values should specify how energy closure and omitted storage terms were handled.
Historical USGS research uses Bowen-ratio energy-balance methods for wetlands and water settings, with explicit attention to the flux ratio, surface energy and water exchange. Such first-party methodological descriptions are more useful than a generic claim that one technique is always most accurate. The suitable method depends on site conditions, observation infrastructure, required time scale and whether the assumptions can be checked.
A two-sentence explanation suitable for a student
First sentence: ‘A Bowen ratio compares the energy warming the air with the energy used to evaporate water; together with measured available surface energy, this ratio helps estimate evapotranspiration.’ Second sentence: ‘The estimate depends on accurate radiation, soil-heat and temperature-humidity measurements and on assumptions about turbulent exchange, so it is not a direct water-meter reading.’ These sentences preserve the mechanism and the limitation without burying a new reader in equations.
Once the reader understands the two sentences, the algebra becomes easier. Write β = H/LE and Rn − G = H + LE, rearrange to LE = (Rn − G)/(1 + β), then convert the time-integrated latent heat through the latent heat of vaporisation into water depth. A reader who understands the units can reproduce a worked example even when variable names change in another textbook. This is what a good technical translation should achieve: transfer of reasoning rather than memorisation of a final formula.
Real scientific communication also needs to distinguish observed, inferred and forecast quantities. Rn and G are measured or estimated flux inputs; β may be inferred from gradients; LE is solved from the adopted balance; and evapotranspiration depth is then calculated from energy using latent heat. If each link is labelled, the explanation remains useful even when a later calibration changes a component of the estimate.
Worked example: estimate one sunny hour of evapotranspiration
An invented field station reports an hourly mean net radiation of 420 W m−2, a downward ground heat flux of 60 W m−2, and a valid Bowen ratio of 0.8 under a convention in which positive energy is available at the surface. Available energy is 420 minus 60 = 360 W m−2. Because β = H/LE, the latent flux is 360 divided by 1.8 = 200 W m−2 and sensible flux is 160 W m−2. Check: 60 plus 200 plus 160 equals 420. This closure check exposes common mistakes before water depth is calculated.
Over one hour, 200 W m−2 transfers 720,000 J m−2 through the latent pathway. Dividing by an illustrative latent heat of 2,450,000 J kg−1 gives 0.29388 kg m−2, approximately 0.294 mm of water-equivalent evapotranspiration over that hour. This figure assumes the flux represents an average over the whole hour and the sign corresponds to upward evaporation. It is not a statement that the site received rain, nor is it a direct measurement of the change in soil-water storage.
To check the result, multiply 0.29388 kg m−2 by 2,450,000 J kg−1 and divide by 3,600 seconds; the answer returns approximately 200 W m−2. If someone instead gets 294 mm, inspect whether they confused kilojoules with joules or kilograms per square metre with grams per square metre. The habit of reversing a calculation catches unit errors that remain hidden in a single forward pass.
Worked example: hourly partition changes across a drying afternoon
Suppose two consecutive one-hour intervals each have 300 W m−2 of available energy. In the first, a valid β of 0.25 gives LE = 300/1.25 = 240 W m−2 and H = 60 W m−2. In the second, β rises to 1.0, giving LE = 300/2 = 150 W m−2 and H = 150 W m−2. Both intervals balance their available energy, but the partition shifts toward warming the air in the second interval.
Using 2.45 MJ kg−1, the two hourly latent fluxes correspond respectively to about 0.353 and 0.220 mm of water. Their total is approximately 0.573 mm for two valid hours. Averaging the two Bowen ratios first would give 0.625 and then a misleading uniform latent flux of about 184.6 W m−2 for both hours, equivalent to about 0.543 mm over two hours. The differing results arise because the equation is nonlinear in β.
The correct conclusion is numerical: the fitted partition changed and the interval-integrated latent estimate declined in the illustration. Explaining why requires more evidence. The site might be drying, vapour-pressure deficit might be changing, wind or cloudiness might differ, or a measurement problem might be developing. A useful scientific report separates this energy calculation from the investigation of its physical cause.
Worked diagnostic: a small vapour gradient creates an enormous Bowen ratio
Assume γ = 0.066 kPa °C−1, ΔT = 0.5 °C and Δe = 0.01 kPa, measured using consistent height order. The simple gradient ratio is β = (0.066 × 0.5)/0.01 = 3.3. Now suppose a humidity-sensor differential offset is only 0.005 kPa. The effective vapour difference could be 0.015 or 0.005 kPa, changing β to about 2.2 or 6.6. The tiny vapour difference and the possible sensor offset are of similar magnitude, which makes the inferred ratio unstable.
If available energy were 250 W m−2, those three ratios would imply latent fluxes of roughly 58.1, 78.1 and 32.9 W m−2 respectively. The arithmetic can be checked, but the spread should not be disguised with excessive decimal precision as though those were three equally observed physical fluxes. The problem is driven by sensitivity of the denominator in a method using small atmospheric gradients. Better differential sensor calibration and quality screening are more important than additional spreadsheet digits.
The repair is to evaluate whether the measured gradient is sufficiently large relative to instrumental uncertainty, examine the direction of the gradient and check whether the two heights sample the same representative surface. Do not replace an unreliable gradient with a convenient nonzero value simply to keep producing an evaporation series. When data cannot support the method’s assumptions, preserve a missing or flagged estimate and use a documented alternative only if justified.
Worked diagnostic: nocturnal heat release and negative energy terms
Imagine a night in which net radiation is negative because a surface loses thermal energy to the sky. The soil or water may release heat accumulated during the day, and turbulent exchange may still occur. If the chosen sign convention labels downward soil heat as positive, soil heat flux can become negative as stored energy moves upward. It is entirely possible for the night-time balance to have combinations of positive and negative terms. Copying a daytime-only formula without retaining signed quantities can reverse the physical story.
Suppose a hypothetical night has Rn = −40 W m−2 and G = −60 W m−2, yielding Rn − G = +20 W m−2. Under a valid measured β = 1, the simplified balance yields LE = 10 W m−2 and H = 10 W m−2. This is not a universal prediction for nights, but it demonstrates that negative net radiation does not automatically mean all turbulent fluxes must be negative: energy can be released from storage. The actual interpretation depends on how all terms were defined and measured.
In another case, vapour may condense at the surface, making the latent exchange negative under an upward-positive convention. That can release energy to other pathways. When a paper uses a different sign convention for condensation or soil heat, translate the definitions before comparing values. A graph with a downward spike should not be labelled ‘negative water use’ until the direction and physical meaning of the latent flux are understood.
An uncertainty budget should include energy closure and representativeness
Consider a simplified estimate with available energy A = 300 W m−2 and β = 0.5, giving LE = 200 W m−2. Suppose the available energy carries a standard uncertainty of 20 W m−2 and β a standard uncertainty of 0.1, treated as independent solely for a teaching calculation. The sensitivity of LE to A is 1/(1+β) = 2/3, giving an A-related contribution around 13.3 W m−2. The sensitivity to β has magnitude A/(1+β) squared = 133.3 W m−2 per unit β, giving another 13.3 W m−2.
Under these invented independence and linearity assumptions, their quadrature combination is about 18.9 W m−2. This is not a verified uncertainty for an actual flux station: wind, surface heterogeneity, differential sensor bias, storage and calibration effects can create correlations or systematic errors outside the calculation. The numerical exercise demonstrates how input sensitivity propagates through an equation rather than granting the resulting figure scientific certainty.
An appropriate field uncertainty assessment asks whether Rn and G are representative of the same footprint as the temperature-humidity gradients, whether a canopy or water body stores material heat, and whether lateral advection transports energy into or out of the analysis region. A mathematically exact closure of an incomplete simplified balance is not evidence that neglected real-world processes vanished. State what was omitted and whether those omissions are likely important for the intended application.
Independent transfer exercises and checked answers
Exercise A: Rn = 500 W m−2, G = 100 W m−2 and β = 1. What are LE and H? Answer: available energy is 400 W m−2, so LE is 400/(1+1) = 200 W m−2 and H = 200 W m−2. Their sum equals 400, and adding the stated G returns the initial 500. A ratio of one means equal partition between sensible and latent components in this model, not that all net radiation goes to water.
Exercise B: γ = 0.066 kPa per degree Celsius, ΔT = 2 °C and Δe = 0.30 kPa. Calculate β. Answer: 0.066 times 2 divided by 0.30 equals 0.44, provided the differences use the same height order and the method’s exchange assumptions are valid. If Δe were accidentally given as 30 percent relative humidity, that computation would be invalid because relative humidity difference is not a vapour-pressure difference.
Exercise C: LE is a valid one-hour average of 150 W m−2, and the latent heat is 2.45 MJ kg−1. Find the hourly evapotranspiration. Answer: 150 times 3,600 divided by 2,450,000 equals approximately 0.220 kg m−2 or 0.220 mm. It would be incorrect to multiply by 86,400 unless 150 W m−2 actually represented an average for a full twenty-four hours.
Exercise D: β is close to −1 and the spreadsheet produces LE greater than 10,000 W m−2. What should the student do? Answer: investigate signs, gradient measurement uncertainty, instrument quality, assumptions and the available-energy terms. The denominator 1+β is close to zero, making the inference numerically unstable. A result should not be published merely because a formula can evaluate it.
Frequently asked questions about Bowen-ratio evaporation
Does a small Bowen ratio always mean a wet landscape? Not necessarily. It indicates a relatively large latent component of the local heat partition when the measurement is valid, but radiation, atmospheric demand, vegetation and other conditions matter. Can a negative Bowen ratio be real? Yes, in certain signed-flux circumstances; however values near minus one can be unstable in the simple energy-balance solution. Is β measured directly? It can be derived from fluxes or inferred from gradients, depending on the method used.
Is sensible heat the same as air temperature? No. A temperature is a state measurement; sensible heat flux describes an energy-transfer rate per area. Is latent heat flux the same as an evaporation depth? Not until it is integrated over time and converted through latent heat of vaporisation. Does evapotranspiration include plants? Yes, the term includes transpiration alongside evaporation pathways, though a single bulk flux method may not separate those components.
What is the most common reporting error? Mixing flux averages, signed terms and time intervals without units. A value in W m−2 is not automatically millimetres per day. Likewise, a daily mean β is not always the correct input for a daily latent-flux estimate. When in doubt, reconstruct the interval energy budget, check the sign definitions and carry units through to the water depth.
A reproducible field-to-reader method for the entire calculation
Begin by identifying the surface and intended footprint, the two sensor heights, wind context and calibration history. Record Rn, G and any storage terms in consistent units and signed conventions. Inspect temperature and actual vapour-pressure gradients and determine how β was estimated. Screen periods with unreliable humidity differences or ratios close to the ill-conditioned region. Only then solve for sensible and latent flux using the appropriate energy budget.
Next integrate valid latent fluxes over the exact intended periods, accounting transparently for missing or screened intervals. Convert joules per square metre to kilograms of evaporated water per square metre using a stated latent heat and then into millimetres. Keep the raw data, calculations, quality flags and any model-based gap filling distinguishable. If the goal is to compare land covers, do not assume the footprints or atmospheric conditions are identical simply because the sites lie in the same region.
Finally communicate in three layers: what was measured, what was calculated from those measurements, and what environmental explanation is supported. For instance, say that net radiation, ground heat and temperature-humidity gradients were observed; a Bowen-ratio method estimated latent flux and interval evapotranspiration; and the partition is consistent with a certain surface response under the documented conditions. This approach makes the numerical result independently teachable and leaves room for evidence that could change its interpretation.
Primary sources and methodological studies
USGS: Evapotranspiration from Upper Klamath Lake, Bowen-ratio energy balance method
USGS: Bowen energy-budget theory and gradient equations
USGS: Evapotranspiration and microclimate field observations
USGS: Bulrush wetland evapotranspiration and seasonal Bowen ratios
Related eduKateSG articles
How to Translate Easily series index
Atmospheric stability, potential temperature and mixing height